apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-04 08:16:42 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

【New Feature】集中式支持w4afp8 (#3644)

Browse Source 
* 支持tp w4afp8

* code style
This commit is contained in:
yangjianfengo1
2025-08-28 10:53:24 +08:00
committed by GitHub
parent 76513f6416
commit e81046fdad
8 changed files with 41 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
custom_ops/gpu_ops/cpp_extensions.cc
View File

@@ -151,7 +151,7 @@ std::vector<paddle::Tensor> MoeExpertDispatch(
const paddle::Tensor &input, const paddle::Tensor &gating_output,
const paddle::optional<paddle::Tensor> &gating_correction_bias,
const paddle::optional<paddle::Tensor> &w4a8_in_scale, const int moe_topk,
const bool group_moe, const bool topk_only_mode);
const bool group_moe, const std::string &moe_quant_type, const bool topk_only_mode);
std::vector<paddle::Tensor>
MoETopKSelectKernel(const paddle::Tensor &gating_logits,
@@ -912,7 +912,7 @@ PYBIND11_MODULE(fastdeploy_ops, m) {
m.def("moe_expert_dispatch", &MoeExpertDispatch, py::arg("input"),
py::arg("gating_output"), py::arg("gating_correction_bias"),
py::arg("w4a8_in_scale"), py::arg("moe_topk"), py::arg("group_moe"),
py::arg("topk_only_mode"), "moe export dispatch function");
py::arg("moe_quant_type"), py::arg("topk_only_mode"), "moe export dispatch function");
/**
* moe/fused_moe/ep_moe_prefill_func.cu